Description of problem: If /boot is on an xfs filesystem, and the directory contains a hole in the extent allocation, booting fails with "error: not a correct xfs inode" Version-Release number of selected component (if applicable): 2.02-142.el8_7.3.x86_64 How reproducible: should be very easy Steps to Reproduce: 1) have older kernel installed 2) create several hundred files with long names in /boot: # for i in {001..300} ; do touch /boot/filler_filename____________________________________${i} done 3) install new kernel version rpms 4) remove the filler files: # rm -f /boot/filler_filename* 5) verify that /boot has a hole: # xfs_bmap /boot boot: 0: [0..7]: 748422104..748422111 1: [8..15]: hole 2: [16..23]: 748422080..748422087 6) reboot Actual results: booting fails with "error: not a correct xfs inode" Expected results: booting succeeds Additional info: bug also affects RHEL 7, but is fixed in RHEL 9
Reproduced using 2.02-142.el8_7.3.x86_64. (wasn't actually that easy to get the hole; I had to try it a few times on different machines) This patch is included in the latest RHEL-8 grub-- 2.02-148.el8-- and therefore in released 8.8. Is this alright for you / the customer or does this need to be backported into specific release(s)? Notably, 8.7.z is already EOL so that patch won't make it there.
Actually, the patch should be in grub2-2.02-123.el8_6.15 (most recent RHEL-8.6 grub) as well.
Marta, I think the patch mentioned in comment#1 is not included in any of the RHEL 8's.I recently checked 8.8 and 8.9 grub(grub2-2.02-148.el8 | grub2-2.02-150.el8) and couldn't find there.However it is included in RHEL 9.I couldn't reproduce the error in recent RHEL 9.3.
Reshmi's totally right in comment#7... not sure what I was seeing before..! Ok, so that means we have a patch and we can do SanityOnly testing since it's difficult to reproduce.
Today's a rough day for me! ;) I have to take back what's in comment#9... I downloaded the src rpm, looked for the patch (xfs: Don't attempt to iterate over empty directory) and didn't find it, but then forgot to apply patches before looking at the code itself. Seems I knew to do this correctly in the past! Turns out (+1 to nfrayer!) that it's in release-to-master.patch So now, also after more info from Nicolas, we think this patch is not enough to fix this issue, especially since it's not very reproducible...